EXAMENSARBETE INOM INFORMATIONS- OCH KOMMUNIKATIONSTEKNIK, AVANCERAD NIVÅ, 30 HP STOCKHOLM, SVERIGE 2016 Content-based Recommender System for Movie Website KE MA KTH SKOLAN FÖR INFORMATIONS- OCH KOMMUNIKATIONSTEKNIK Content-based Recommender System for Movie Website KE MA Master’s Thesis at VionLabs Supervisor: Chang Gao Examiner: Mihhail Matskin TRITA xxx yyyy-nn Abstract Recommender System is a tool helping users find content and overcome information overload. It predicts interests of users and makes recom- mendation according to the interest model of users. The original content-based recommender system is the continuation and development of collaborative filtering, which doesn’t need the user’s evaluation for items. Instead, the similarity is calculated based on the information of items that are chose by users, and then make the rec- ommendation accordingly. With the improvement of machine learning, current content-based recommender system can build profile for users and products respectively. Building or updating the profile according to the analysis of items that are bought or visited by users. The sys- tem can compare the user and the profile of items and then recommend the most similar products. So this recommender method that compare user and product directly cannot be brought into collaborative filtering model. The foundation of content-based algorithm is acquisition and quantitative analysis of the content. As the research of acquisition and filtering of text information are mature, many current content-based recommender systems make recommendation according to the analysis of text information. This paper introduces content-based recommender system for the movie website of VionLabs. There are a lot of features extracted from the movie, they are diversity and unique, which is also the difference from other recommender systems. We use these features to construct movie model and calculate similarity. We introduce a new approach for setting weight of features, which improves the representative of movies. Finally we evaluate the approach to illustrate the improvement. Key words: recommender system, content-based, collabo- rative filtering, similarity, movie Referat Recommender System är ett verktyg som hjälper användarna att hitta innehåll och övervinna informationsöverflöd. Det förutspår användarnas intressen och gör rekommendation enligt räntemodellen användare. Den ursprungliga innehållsbaserade recommender är en fortsättning och utveckling av samarbete filtrering, som inte behöver användarens ut- värdering artiklar. Istället är likheten beräknas baserat på informationen objekt som har varit valde av användare, och sedan göra rekommenda- tionen därefter. Med förbättringen av maskininlärning, kan nuvarande innehållsbaserad recommender systemet bygga profil för användare och produkt respektive. Bygga eller uppdatera profilen enligt analysen av objekt som köps eller besöks av användare. Systemet kan jämföra an- vändaren och profilen av artiklar och rekommendera den mest liknande produkt. Så här recommender metod som jämför användaren och pro- dukten direkt kan inte föras in collaborative filtreringsmodell. Grunden för innehållsbaserad algoritm är förvärv och kvantitativ analys av inne- hållet. Eftersom forskning förvärv och filtrering av textinformation är mogen, många aktuella innehållsbaserade recommender system gör re- kommendation enligt analysen av textinformation. Denna uppsats införa innehållsbaserad recommender system för film webbplats VionLabs. Det finns en mängd funktioner som extraherats från en film, är de mångfald och unik, vilket är också skillnaden med andra recommender system. Vi använder dessa funktioner för att kon- struera film vektor och beräkna likheter. Vi introducerar en ny metod för att fastställa vikten av funktioner, vilket förbättrar företrädare för filmer. Slutligen utvärderar vi tillvägagångssättet för att illustrera för- bättringen. Nyckelord: recommender system innehållsbaserad, samar- bets filtrering, likheten, film Contents 1 Introduction 1 1.1 Background . 1 1.2 Problem Statement . 2 1.3 Goals . 2 1.4 Methodology . 2 1.5 Ethics . 3 1.6 Delimitations . 3 1.7 Outline . 3 2 Overview of Related Work 5 2.1 Content-based Recommendation . 5 2.2 Collaborative-filtering Recommendation . 7 2.2.1 User-based collaborative-filtering . 7 2.2.2 Item-based collaborative-filtering . 8 2.3 Hybrid Recommender Systems . 9 2.4 Comparison . 9 2.5 Famous Recommender Systems . 10 2.5.1 E-Commerce . 10 2.5.2 Movie and Video website . 10 2.5.3 Internet Radio . 14 3 Recommender System for Vionel 15 3.1 Datasets . 15 3.2 Feature Extraction . 15 3.2.1 Singular Value Decomposition . 16 3.2.2 Matrix Low Rank Approximation . 16 3.2.3 Application of LSI . 17 3.3 Feature Representation . 17 3.4 TF-IDF . 18 3.4.1 Term Frequency . 18 3.4.2 Inverse Document Frequency . 18 3.4.3 Normalization . 19 3.5 Weakness of TF-IDF . 19 3.6 Improvement of TF-IDF . 19 3.6.1 Improved IDF . 20 3.6.2 Distribution Coefficient . 21 3.6.3 Normalization . 22 3.7 Similarity . 22 4 Experiment 25 4.1 Dataset . 25 4.2 Category . 26 4.3 Document . 26 4.4 Result . 28 5 Validation 31 5.1 k-NN . 31 5.2 Evaluation . 32 5.3 Evaluation Metrics . 32 5.4 Analysis . 33 6 Conclusion 35 7 Future Work 37 Bibliography 39 Chapter 1 Introduction As the development of information technology and Internet, people enter the era of information overload from that of information deficiency gradually[17]. The report is the result of the Master Thesis in Information and Communication Technology School at Royal Institute of Technology in Stockholm, Sweden. The project is provided in VionLabs AB that is a media-tech company providing media content such as movie information for customers. The recommender system I implemented is for the movie website Vionel.com. 1.1 Background In the era of information overload, it is very difficult for users to get information that they are really interested in. And for the content provider, it is also very hard for them to make their content stand out from the crowd. That is why many re- searchers and companies develop Recommender System to solve the contradiction. The mission of Recommender System is to connect users and information, which in one way helps users to find information valuable to them and in another way push the information to specific users. This is the win-win situation for both customers and content providers. VionLabs is a media-tech startup company. The company provides a new way on how consumers are given access to good and suitable content. The mission of VionLabs is to increase needs of its digital user base. Vionel is the movie website developed by VionLabs, which is a place for people who love movies can gather all the information about films in one place[5]. This thesis report will present a more practical recommendation method that can be used on a movie website that does not have enough users. 1 CHAPTER 1. INTRODUCTION 1.2 Problem Statement For building a recommender system from scratch, we face several different problems. Currently there are a lot of recommender systems based on the user information, so what should we do if the website has not gotten enough users. After that, we will solve the representation of a movie, which is how a system can understand a movie. That is the precondition for comparing similarity between two movies. Movie features such as genre, actor and director is a way that can categorize movies. But for each feature of the movie, there should be different weight for them and each of them plays a different role for recommendation. So we get these questions: • How to recommend movies when there are no user information. • What kind of movie features can be used for the recommender system. • How to calculate the similarity between two movies. • Is it possible to set weight for each feature. 1.3 Goals The goals of this thesis project is to do the research of Recommender Systems and find a suitable way to implement it for Vionel.com. There are many kinds of Recommender Systems but not all of them are suitable for one specific problem and situation. Our goal is to find a new way to improve the classification of movies, which is the requirement of improving content-based recommender systems. 1.4 Methodology In order to achieve the goal of the project, the first process is to do enough back- ground study, so the literature study will be conducted. The whole project is based on a big amount of movie data so that we choose quantitative research method. For philosophical assumption, positivism is selected because the project is experi- mental and testing character. The research approach is deductive approach as the improvement of our research will be tested by deducing and testing a theory. Ex post facto research is our research strategy, the movie data is already collected and we don’t change the independent variables. We use experiments to collect movie data. Computational mathematics is used data analysis because the result is based on improvement of algorithm. For the quality assurance, we have a detail explana- tion of algorithm to ensure test validity. The similar results will be generated when we run the same data multiple times, which is for reliability. We ensure the same data leading to same result by different researchers.[12] 2 1.5. ETHICS 1.5 Ethics Movie information is the only part that may have ethics problem. However, all the information we get for research is from public database such as Wikipedia and our own movie database. So there are no data confidentiality and user privacy problems. 1.6 Delimitations In this project, we will not deep in how to collect and generate the data, which is the business of other teams in VionLabs. In the validation part, the data is divided into training and testing parts. But how to split data will not presented in detail neither. It is important to build a real recommender system for a business company and in the end I have built a successful demo which helped our company to get the opportunity of cooperation with other companies.